By buld_online 
Google has quietly released a new experimental app called AI Edge Gallery, allowing Android users to run a variety of open-source artificial intelligence (AI) models directly on their devices, no internet connection required.
Available now as an alpha release on GitHub, AI Edge Gallery lets users download and run compatible AI models for tasks like image analysis, chat-based interactions, text generation, and code editing. These models run entirely on-device, tapping into the phone’s CPU, GPU, or other accelerators, making AI more accessible in offline environments and reducing privacy concerns associated with cloud-based processing.
The app, which is not yet available on the Google Play Store and must be sideloaded, provides access to open-source models from platforms like Hugging Face, including Google’s recently announced Gemma 3n. Once installed, these models can help analyse images, answer queries, rewrite text, and generate code, all while operating offline, even in airplane mode.
AI Edge Gallery features three main tools:
- Ask Image: Upload a photo and ask questions about its content.
- AI Chat: Interact with a model in a chat format.
- Prompt Lab: Perform tasks like text summarisation, tone-based rewriting, or code snippet generation using configurable AI prompts.
The performance of these models depends heavily on your device’s hardware. Larger models and more complex tasks require more powerful chips, meaning older or mid-range phones may struggle with speed or thermal efficiency. Google allows users to choose which processor (CPU, GPU, etc.) the model runs on, monitor temperature, and adjust configuration settings like tokens and temperature values. Battery consumption and overheating could become concerns during extended use, especially on less capable devices.
Google’s move underscores its broader vision of bringing AI closer to the edge, on personal devices rather than relying solely on cloud computing. While the app’s on-device models may not yet rival the capabilities of full-scale tools like ChatGPT or Gemini, the effort marks a significant step towards more private, responsive, and portable AI experiences.
The AI Edge Gallery app is currently available under the permissive Apache 2.0 licence, which allows both academic and commercial use. Although currently limited to Android users, Google has confirmed that an iOS version is in development.
Developers and early adopters interested in experimenting with local AI capabilities can download the APK and follow installation instructions via Google’s GitHub page. However, those unfamiliar with sideloading may want to wait for the app’s official release on the Play Store for added security and convenience.